thesis-engine

Thesis Engine — Research Feeder for vibe-thesis

A three-stage engine: Discover → Gather → (optional) Adapt.

The engine's job is to do the cold-start work that's painful inside vibe-thesis: scanning the field for live topics, pulling primary sources and opposing positions, and shaping them into vibe-thesis-compatible notes and a BibTeX bibliography. The drafting itself stays inside vibe-thesis where the persona and citation pipeline live.

---

What this skill is NOT

• It is not a thesis writer. The full-paper Stage 2 from v0.1 is gone — vibe-thesis's 03_BODY/ chapter scaffolding plus the LeadWriter persona handle the actual drafting.
• It is not a citation manager. It produces a BibTeX file. Merging into your canonical 05_CITATIONS/references.bib is a manual or scripted step.
• It does not modify a live ThesisStudio. Output always lands in a dated run folder; the human (or a future ingest skill) copies subdirs across.

---

Stage 1 — Topic Discovery

Goal: Surface 5 high-signal, low-competition thesis topics via web search.

Step 1.1 — Domain Selection

Default domain is ai_ml. Honor explicit --domain arguments. Supported domains:

• ai_ml — AI / Machine Learning (default)
• theater_ops — Theater Technology & Operations
• film_exhibition — Film Distribution & Exhibition
• spatial_xr — Spatial Computing / XR
• behavioral_econ — Behavioral Economics
• org_behavior — Organizational Behavior
• data_eng — Data Engineering
• Any custom domain string

Step 1.2 — Search for Cutting-Edge Topics

Run 3–5 web searches across these angles:

[domain] research breakthroughs 2025 2026
[domain] unsolved problems emerging trends
[domain] preprint arxiv notable 2025
[domain] industry disruption underreported

For each result cluster, extract:

• Topic name (5–10 words)
• Novelty score (1–5): Is this still emerging vs already mainstream?
• Thesis potential (1–5): Is there an arguable, researchable angle?
• Relevance (1–5): Connects to the user's domain mix?

Step 1.3 — Present Top 5

Show a ranked table:

#	Topic	Novelty	Thesis Potential	Relevance	Total
1	...	4	5	4	13

Ask the user to pick one, or auto-select the top scorer if --auto is passed.

Write the table to <run>/topics.md so it survives the session.

---

Stage 2 — Source Gathering

Goal: For the chosen topic, produce vibe-thesis-shaped research notes and a starter BibTeX bibliography.

Output structure

thesis-engine-run-YYYYMMDD/
├── README.md                       (run summary + ingest instructions)
├── topics.md                       (Stage 1 ranking)
├── 01_PLANNING/
│   └── proposal.md                 (thesis proposal stub: what, why, scope)
├── 02_RESEARCH/
│   ├── prior-art/notes.md
│   ├── methodology-survey/notes.md
│   ├── opposing-positions/notes.md
│   ├── key-authors/notes.md
│   └── primary-sources/notes.md
└── 05_CITATIONS/
    └── references.bib              (BibTeX, Pandoc-ready)

The subdirectory names mirror the vibe-thesis Template exactly — drop them into C:\Users\estev\Projects\ThesisStudio\ and they merge cleanly. references.bib should be appended to the canonical 05_CITATIONS/references.bib (or replace it on a fresh project).

Step 2.1 — Source Search

Run 5–8 targeted web searches structured by axis:

Axis	Searches	Target output
Prior art	"[topic] literature review", "[topic] survey 2024 2025"	3–5 sources
Methodology	"[topic] methodology", "how researchers measure [topic]"	2–3 sources
Opposing	"[topic] critique", "[topic] limitations", "against [topic]"	2–3 sources
Key authors	"[topic] researchers", "leading [topic] scholars"	2–4 named authors
Primary	preprints, datasets, original papers	2–3 sources

For each source, capture:

• Title
• Author(s)
• Year
• Venue (journal, conference, blog, preprint server)
• URL
• 2–3 sentence summary
• Pandoc citation key (authoryear style — e.g., smith2024)

Step 2.2 — Write Research Notes

Each axis gets its own notes file with this template:

# [Axis Name] — [Topic]

## Summary
[2–3 sentence framing of what this axis tells us about the topic]

## Sources

### [@citation-key] — [Title]
- **Author(s)**: ...
- **Year**: ...
- **Venue**: ...
- **URL**: ...
- **Why it matters**: 2–3 sentences specific to the thesis.
- **Quote / data point**: optional pull-quote.

### [@next-key] — ...

Use Pandoc citation keys throughout. Body claims should reference sources as [@key] so a future drafting step can lift them straight into 03_BODY/.

Step 2.3 — Write the Proposal Stub

01_PLANNING/proposal.md:

# [Topic Title] — Thesis Proposal (Engine-Generated Stub)

> Generated by Thesis Engine on YYYY-MM-DD.
> Edit freely in vibe-thesis. The engine fills in scaffold; the human owns the argument.

## What this thesis argues
[1–2 sentence claim derived from Stage 1 scoring rationale]

## Why it matters
[Why now? What's the gap in the field? Pull from prior-art notes.]

## Scope
- **In scope**: ...
- **Out of scope**: ...

## Methodology sketch
[Pull from methodology-survey notes — what approach do similar studies use?]

## Anticipated counterarguments
[Pull from opposing-positions notes — name them now so vibe-thesis can address them in-text.]

## Next steps inside vibe-thesis
1. Decide `THESIS_MODE` (dissertation / article / masters).
2. Drop `02_RESEARCH/` and `05_CITATIONS/references.bib` into ThesisStudio.
3. Refine claim → run a research swarm via `04_AGENT_SWARMS/` if deeper coverage is needed.
4. Begin drafting in `03_BODY/`.

Step 2.4 — Write the BibTeX File

05_CITATIONS/references.bib:

@article{smith2024,
  title   = {Title here},
  author  = {Smith, Jane},
  year    = {2024},
  journal = {Venue Name},
  url     = {https://...},
  note    = {Engine-gathered. Verify before citing.}
}

@inproceedings{jones2025,
  ...
}

Cite type rules: @article for journal/blog, @inproceedings for conference, @misc for preprints/datasets/web sources without a clear venue, @book for books.

The note = {Engine-gathered. Verify before citing.} line is intentional — it surfaces in rendered bibliographies and reminds the human to verify each entry before relying on it. Strip the note line after verification.

Step 2.5 — Quality Gates

Before completing:

• At least 12 sources total across all axes (3+ prior art, 2+ opposing, others as found)
• Each source has a Pandoc citation key
• Each notes file references its sources with [@key] syntax
• references.bib parses (no malformed entries)
• proposal.md has a single explicit thesis claim
• Each 02_RESEARCH/<axis>/notes.md has a Summary section

If any gate fails, surface the failed gates as a checklist in chat and write the partial run to <run>/pending/. Do not silently proceed.

---

Stage 3 — Blog Adaptation (Optional)

Goal: From the Stage 2 proposal + research notes, produce a Smart Brevity blog draft for BlogStudio. This stage is optional and only runs when invoked via /thesis-engine:blog.

Output structure

thesis-engine-run-YYYYMMDD/
└── blog/
    └── 02_DRAFTS/
        └── YYYY-MM-DD-[slug]/
            ├── POST.md
            └── frontmatter.yaml

Subdirectory names mirror BlogStudio's live layout — drop 02_DRAFTS/<dated-slug>/ into C:\Users\estev\Projects\BlogStudio\02_DRAFTS\ for ingest.

Step 3.1 — Distillation Rules

Source material	Blog equivalent
proposal.md claim	Hook paragraph + bold one-line headline
Prior-art notes	"What we thought we knew" (2–3 paragraphs)
Methodology notes	Brief — only if methodology is the story
Opposing-positions notes	"The strongest case against this" mini-section
Primary sources	Pull-quote or data callout

Step 3.2 — Blog Post Specs

• Length: 800–1,200 words
• Tone: Smart Brevity — punchy, scannable, no jargon, lead with the bottom line
• Format: ## headers, paragraphs ≤3 sentences, 1–2 pull-quote callouts, bullet lists
• Title: ≤60 characters, curiosity-gap or claim style
• Tags: 3–5 keywords from topic + axes
• Citations: Inline link out (no Pandoc syntax — this is a blog draft, not a paper)

Step 3.3 — frontmatter.yaml

title: "..."
slug: "kebab-case-title"
excerpt: "1-2 sentence teaser, ≤140 chars"
tags: [tag1, tag2, tag3]
domain: "..."
source_run: "thesis-engine-run-YYYYMMDD"
generated_date: "YYYY-MM-DD"
status: draft

---

Repeatable Run Protocol

When the user invokes /thesis-engine:run, execute Stages 1 + 2 in sequence. Stage 3 is opt-in via --blog or via the standalone /thesis-engine:blog command.

After a successful run:

1. Write <run>/README.md summarizing what was produced and how to ingest it into ThesisStudio.
2. Log a completion task to the 626Labs Thesis Engine project (project ID 6H02m934H97jNl15DzWe) using mcp__626labs__manage_tasks action create with title "Thesis run YYYY-MM-DD — [topic]" and status Done. If the 626Labs MCP is unavailable, log a one-line note to chat and continue.
3. List the output paths in chat with computer:// links.

---

Recurring Cadence Options

Frequency	Trigger	Output
Weekly	Scheduled task weekly-thesis-engine (Mondays 8 AM, --domain ai_ml --auto)	1 topic + research bundle, no blog
On-demand	/thesis-engine:run	Same
Batch	/thesis-engine:run --count 3	3 topics + 3 research bundles
Discover-only	/thesis-engine:discover	Topic table only
Blog-only	/thesis-engine:blog [run-folder-or-thesis.md]	Blog draft from existing run or thesis

---

Integration Notes

• vibe-thesis (ThesisStudio) — 01_PLANNING/, 02_RESEARCH/, 05_CITATIONS/ from a run folder drop directly into C:\Users\estev\Projects\ThesisStudio\. Subdirectory names match exactly. Merge references.bib into the canonical bibliography.
• BlogStudio — blog/02_DRAFTS/<dated-slug>/ drops into C:\Users\estev\Projects\BlogStudio\02_DRAFTS\.
• 626Labs MCP — Each run logs a completion task to project 6H02m934H97jNl15DzWe.
• Citation style — Pandoc [@authorYear] keys throughout, BibTeX bibliography. Matches vibe-thesis's render pipeline.

---

References

See references/domain-feeds.md for a curated list of RSS feeds, preprint servers, and newsletters to seed Stage 1 discovery searches.